Japanese enterprises deploy lightweight LLM tsuzumi 2 on single GPU, achieving frontier model performance while cutting infrastructure costs and addressing data sovereignty concerns.